atlas · the context layer

What your company knows, ready for any agent.

Atlas holds the skills, facts and metric definitions your company agrees on. Every item has a version, a source and a scope. Any agent can read it, and an agent only sees what the people in the room can see.

Three kinds of context. Any number of readers.

Skills are how work gets done. Knowledge is what is true. Metrics are what words mean. Each is kept once, in Atlas, and every agent reads the same copy.

Context graph

what Atlas holds · who reads it

one thread in #financeeverything
skillsskills/month-end-close.mdv14 · team/finance · confirmedskills/deploy-runbook.mdv22 · team/engknowledgepayouts run on the 28thsource #finance · team/financerefunds over $2k need a sign-offsource thread · team/financemetricsmetrics/churn.sqllogos lost / period · v2 · financemetrics/net_revenue_retention.sqlv3 · company · source billingMaia in #financethe team's coworker · nativeClaude Codereads the runbooks over MCPan OpenAI agentsame scope rules, over MCPyour own agentany model, over MCPa personconsole and CLIscope · version · source

Everything Atlas keeps, and how it keeps it.

Pick a type of context to see what it looks like inside Atlas, or a mechanism to see how it is kept honest.

01

Skills

How a job is done here. A skill is a procedure with a trigger, kept as versions. A correction confirmed in a thread becomes the next version.

skill · month-end-close · v14 · team/finance

02

Knowledge

What is true here. Decisions, facts and definitions, each with the source it came from and the scope that may see it.

entry · payouts run on the 28th · source #finance

03

Metrics

What a word means here. A metric is agreed once, with its grain and its source, and every agent answers from the same definition in every channel.

metric · churn = logos lost / period · v2

04

Scopes

Who may see it. Every object belongs to the company, a team or a person. An agent in a shared thread only uses what everyone there can see.

scope · company › team/finance › person/d.reyes

05

Versions

Every change is a new version with who made it and where. Nothing is overwritten. Roll back in one step.

v11 › v12 › v13 (corrected) › v14 (live)

06

Confirmation

An agent proposes, a teammate confirms, in the thread where the work happened. Only confirmed changes become versions.

proposed by maia · confirmed by d.reyes · 9:41

07

Readers

Maia reads Atlas natively. Claude Code, an OpenAI agent or your own agent reads it over MCP, under the same scope rules.

protocol · MCP · scope enforced on every read

08

Console and CLI

People edit, confirm and revert from a browser or a terminal. What an agent sees, a person can open.

atlas skills show month-end-close --version 14

09

Single-tenant

One deployment per customer, in your cloud or ours. Model calls go to the providers you approve, on your keys. Nothing else leaves.

1 tenant · your keys · approved providers only

10

Audit log

Every read and every change is logged: who, what, which scope, which version. Exportable.

read · maia · month-end-close v14 · #finance · 9:41

Start with one team.

Bring one team's procedures, facts and definitions. We run Atlas for you and connect the agents you already use.